Hi all, Pierre,

Here's the much smaller patch I mentioned at gdb@.

This version is about as small as it can get.  First the
prologue is analysed and skipped, then line info is used
to get at the first line of user code in the function.  If
the function is called "main" (humm, I shall move that into
the callback, as I believe the place where __main/__gccmain
is emitted is configurable on the gcc side), call a gdbarch
method that skips the __main call.  Works OK as long as nothing
is scheduled into before the __main call.  That means that it
should work OK at -O0, which is enough for fixing the
runto_main problems in the testsuite.  Then again,
debugging code with .stabs debug info at anything but -O0
is not pleasant anyway.

Last time I looked at the results, it uncovered other
more serious problems, so it was a win just for that fact.

I'm starting an overnight testrun on Cygwin, as the
last time I did that was a couple of months back.

Any comments?  Suggestions for a better gdbarch
callback name?

-- 
Pedro Alves